Joint image segmentation method

Abstract

The invention discloses a joint image segmentation method, which comprises the following steps of: S1, acquiring an image containing a joint to be segmented, judging whether metal exists in the image or not, and removing a metal area in the image when the metal exists; s2, acquiring a skeleton region image in the image obtained in the S1; s3, performing threshold segmentation on the skeleton region image obtained in the S2 to obtain masks corresponding to skeletons in the skeleton region image, and performing segmentation by adopting a graph segmentation algorithm to obtain each skeleton in the skeleton region image to obtain a skeleton segmentation image; s4, calculating the distance between each point in the skeleton segmentation image obtained in the step S3 and the nearest boundary thereof to obtain a distance map; obtaining a seed region of each skeleton according to the characteristics of the skeleton region in the joint, and marking to obtain a seed region label map; and performing semantic segmentation on the skeleton segmentation image obtained in the step S3 through a watershed algorithm. According to the method, joint image segmentation can be realized under the adhesion condition, and defects such as cavities do not occur in the segmented image.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a joint image segmentation method. Background Art

[0002] The surface of human joints, such as the knee joint, is covered with cartilage. As we age, the cartilage gradually wears away, and the bones and ligaments around the joint also degenerate, eventually causing joint pain, deformity, and movement disorders. Therefore, joint replacement surgery, such as knee replacement surgery, has become an important way to solve this medical problem. However, many people have a misunderstanding that joint replacement surgery means replacing the entire joint, but this is not the case. Joint replacement surgery mainly replaces the cartilage on the surface of the joint that has been worn into pits with metal prostheses and wear-resistant gaskets made of high-molecular polyethylene.

[0003] Taking the knee joint as an example, before the replacement surgery, the patient's knee joint needs to be imaged by CT 3D to divide the femur, tibia and fibula of the knee joint as accurately as possible. Therefore, high-quality image semantic segmentation plays a vital role in medical image analysis.

[0004] However, there are two difficulties in joint medical image segmentation:

[0005] (1) The boundaries between the bones on both sides of the joint and the cartilage in the middle are blurred, especially in patients with severe joint injuries. The cartilage is severely worn and there is no obvious dividing line under imaging, which is often called adhesion.

[0006] (2) There is often low-density tissue around the joints of the human body. The images segmented by common image segmentation methods often have defects such as holes, or cannot effectively segment bones and muscles. Summary of the invention

[0007] Purpose of the invention: In view of the above-mentioned deficiencies, the present invention proposes a joint image segmentation method, which can realize joint image segmentation in the case of adhesion, and defects such as holes will not appear in the segmented images.

[0008] Technical solution:

[0009] The present invention provides a joint image segmentation method, comprising:

[0010] S1, obtaining an image containing a joint to be segmented, determining whether there is metal therein, and removing the metal area therein if there is metal;

[0011] S2, obtaining a bone region image from the image obtained in S1;

[0012] S3, performing threshold segmentation on the bone region image obtained in S2 to obtain the mask corresponding to the bone, based on which, a graph cut algorithm is used to segment each bone therein, thereby obtaining a bone segmentation image;

[0013] S4, calculating the distance between each point in the bone segmentation image obtained in S3 and its nearest boundary to obtain a distance map;

[0014] According to the characteristics of the bone region in the joint, the seed region of each bone is obtained, and the seed region label map is obtained by marking;

[0015] The skeleton segmentation image obtained by S3 is semantically segmented using the watershed algorithm.

[0016] Specifically, in S3, the step of segmenting the skeleton using a graph cut algorithm includes:

[0017] S31, creating a priori graph model according to the mask corresponding to the obtained skeleton;

[0018] S32, solving the graph model by using a minimum cut or maximum flow algorithm to obtain a bone segmentation result, that is, segmenting each bone in the bone region image.

[0019] More specifically, in S31, the graph model is as follows:

[0020] The graph model is specifically as follows:

[0021] Fill the closed contour of the mask corresponding to the bone on the corresponding view, use it as the foreground area, and mark the voxel value of the voxel therein as 2;

[0022] According to S2, the area outside the bone area of ​​the mask corresponding to the bone on the corresponding view is used as the determined background area, and the voxel value of the voxel therein is marked as 1;

[0023] The other areas in the mask corresponding to the bones on the corresponding view are regarded as unknown areas, and the voxel value of the voxels in the marker is 0;

[0024] Thus, a priori graph model is created, where S represents the foreground, T represents the background, and undirected edges are established between the elements in the graph model;

[0025] For voxels with a marked voxel value of 1, set their weight to S to infinity and their weight to T to 0; for voxels with a marked voxel value of 2, set their weight to S to 0 and their weight to T to infinity; for voxels with a marked voxel value of 0, set their weight to S to -λ*In(I / Hist B ), the weight to T is -λ*Ln(I / Hist o ), where λ is the set weight coefficient, I is the voxel value of the current voxel, HistO The weight of the voxel with a marked voxel value of 0 to the foreground voxel, Hist B is the weight of the voxels with a marked voxel value of 0 to the background voxels.

[0026] Specifically, in S3, after the graph cut algorithm is used to segment the bones, connected domains can be screened as follows: calculate and obtain the connected domain with the largest area among the connected domains corresponding to each bone, and retain the connected domains corresponding to the areas greater than 0.5 times the maximum area as the bones in the bone area image obtained by the final segmentation.

[0027] Specifically, in S4, the marking obtains a seed region label map, which is as follows:

[0028] (1) Find the joint gap position in the bone segmentation image obtained by S3;

[0029] (2) according to the distance map obtained by calculation, the minimum value of the distance therein is multiplied by a set coefficient to obtain a set distance threshold, the bone segmentation image is segmented according to the set distance threshold, and a plurality of connected domains are obtained through connected domain analysis;

[0030] (3) Calculate the size of each connected domain in step (2), determine the corresponding bone region according to the joint gap position obtained in step (1), and mark it as the corresponding seed region, thereby obtaining a seed region label map.

[0031] More specifically, in step (1), the step of finding the joint gap position in the bone segmentation image is as follows:

[0032] Performing maximum density projection on the bone segmentation image along the left-right direction of the human body to obtain a corresponding two-dimensional image, and performing threshold segmentation on the two-dimensional image to obtain a binary image;

[0033] The pixel values ​​of the pixel points in the same column along the front-back direction of the human body in the binary image are accumulated to obtain a corresponding pixel value accumulation value curve, and the average value of the pixel values ​​of all the pixel points in the binary image is used as a reference value, and the area in which the accumulated values ​​of two continuous and smooth pixel values ​​are lower than the reference value is obtained, which is the joint area;

[0034] The pixel points in the column corresponding to the minimum pixel value cumulative value in the pixel value cumulative value curve in the joint area are obtained, and the pixel points in this column are the corresponding joint gap positions.

[0035] Specifically, the S2 includes:

[0036] S21, perform maximum density projection on the image obtained in S1 along the direction of the head and feet of the human body, perform threshold segmentation on the obtained projection image, and then perform connected domain operation to obtain several connected domains, obtain the circumscribed rectangle of the connected domain with the largest area as a bounding box, and use the bounding box to select the image obtained in S1 to obtain the region of interest in the image;

[0037] S22, performing maximum density projection on the region of interest obtained in S21 along the direction of the head and feet of the human body, performing threshold segmentation on the obtained projection image, obtaining its maximum circumscribed rectangle, and cropping the image of the bone region therein based on the maximum circumscribed rectangle.

[0038] More specifically, in S21, the threshold segmentation of the projection image of the image adopts the moments histogram threshold segmentation algorithm.

[0039] More specifically, in S22, the projection image of the region of interest is threshold segmented using an OTSU threshold segmentation algorithm.

[0040] Specifically, in S1, the determination of whether there is metal in the image of the joint to be segmented is as follows:

[0041] After using threshold segmentation to segment the cross-sectional, sagittal and coronal views of the image respectively, the mean Hu value of the foreground area is statistically obtained. If the mean is greater than the set threshold, it is considered that there is metal in the image, otherwise there is no metal in the image.

[0042] Specifically, in S1, the metal area in the image containing the joint to be segmented is removed, specifically:

[0043] The Hu values ​​of the points in the metal area of ​​the image containing the joint to be segmented are replaced with the set value.

[0044] Beneficial effects: The present invention uses threshold segmentation to provide prior foreground and background knowledge based on the graph cut algorithm, obtains the mask of the bone part through calculation, and segments the bone area of ​​the foreground. It can realize joint image segmentation in the case of adhesion, and there will be no defects such as holes in the segmented image. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are merely embodiments of the present invention, and ordinary technicians in this field can obtain other drawings based on these drawings without creative work.

[0046] Figure 1 is a flow chart of the joint image segmentation method of the present invention;

[0047] Figure 2 This is an example of an image containing a joint obtained by box selection, where: Figure 2 (a) is an example of selecting a frame in a human image. Figure 2 (b) is an example of an image containing a joint obtained by box selection;

[0048] Figure 3 This is an example of obtaining an image of the bone region in an image, where: Figure 3 (a) is an example of a projection image obtained by maximum density projection along the direction of the head and feet of the human body. Figure 3 (b) is an example diagram for obtaining the bounding rectangle of the connected domain with the largest area in the projection image;

[0049] Figure 4 This is an example of the maximum bounding rectangle obtained in the region of interest;

[0050] Figure 5 This is an example of the result obtained by threshold segmentation of the bone region image histogram;

[0051] Figure 6 This is an example of the result obtained by filling the closed contour of the mask corresponding to the bone;

[0052] Figure 7 An example graph of the established prior graphical model;

[0053] Figure 8 This is an example of the result obtained by segmenting the bone area image using the graph cut algorithm;

[0054] Fig. 9 is an example of the distance map calculated;

[0055] Fig.10 An example diagram for finding the location of joint gaps in bone segmentation images, where: Fig.10 (a) is an example image obtained by performing maximum density projection and binarization on the bone segmentation image along the left and right directions of the human body; Fig.10 (b) is an example diagram of the obtained pixel value accumulation value curve;

[0056] Fig.11 This is an example of the obtained seed region label map;

[0057] Fig.12 Flow chart for semantic segmentation of bone segmentation images using watershed algorithm;

[0058] Fig.13 This is an example of the result of semantic segmentation.

[0059] The exemplary images of the present invention generally include a cross-sectional view, a sagittal view, a coronal view and a three-dimensional image. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0061] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present invention should have the common meanings understood by persons having ordinary skills in the field to which the present invention belongs.

[0062] The process of the joint image segmentation method of the present invention is as follows Figure 1 As shown, including:

[0063] S1, obtaining an image containing a joint to be segmented, determining whether there is metal therein, and removing the metal area in the image if there is metal;

[0064] In the present invention, a human body image is obtained, the joint area to be segmented is framed out, and then the image containing the joint is obtained by cutting, such as Figure 2 As shown; in this embodiment, the joint to be segmented is the knee joint. Of course, it can also be an image containing human joints obtained from other means, such as an image containing joints obtained through AI recognition.

[0065] In the present invention, to determine whether there is metal in the image, after segmenting the image using threshold segmentation, the Hu value average of the foreground area is statistically obtained. If the average is greater than a set threshold, it is considered that there is metal in the image, otherwise there is no metal in the image. In this embodiment, the threshold segmentation of this step can adopt the Yen histogram threshold segmentation algorithm.

[0066] In the present invention, the threshold value may be set to a value greater than 1500, and in this embodiment, the threshold value is 2000.

[0067] In the present invention, if there is metal in the image, the metal region can be removed by replacing the Hu value of the points in the metal region with a set value. In this embodiment, the set value can be the Hu value of air, that is, -1000, or -1024.

[0068] S2, obtaining a bone region image from the image obtained in S1;

[0069] In the present invention, obtaining the bone region image in the image obtained in S1 includes:

[0070] S21, perform maximum density projection on the image obtained in S1 along the direction of the head and feet of the human body, perform threshold segmentation on the obtained projection image, and then perform connected domain operation to obtain several connected domains, obtain the circumscribed rectangle of the connected domain with the largest area as a bounding box, and use the bounding box to select the image obtained in S1 to obtain the region of interest in the image;

[0071] In the present invention, the obtained projection image is as follows Figure 3 As shown in (a), the obtained region of interest is Figure 3 (b) as shown.

[0072] In the present invention, the window position and width of the image obtained by S1 can be adjusted first, and then the corresponding projection image can be obtained; in this embodiment, the bone window can be set to [-400, 850].

[0073] In the present invention, the threshold segmentation performed on the projection image may adopt the moments histogram threshold segmentation algorithm.

[0074] In the present invention, after the projection image is segmented by threshold value, a closing operation may be performed on it, and then a connected domain operation may be performed on it, thereby obtaining a plurality of more accurate connected domains.

[0075] In the present invention, if the image obtained by S1 is an image after the metal area is eliminated, the boundary box obtained in this step can be expanded by 10 pixels on the basis of the image obtained by S1, but cannot exceed the range of the image obtained by S1, so as to reduce the amount of calculation while eliminating the influence of metal.

[0076] S22, perform maximum density projection on the region of interest obtained in S21 along the direction of the head and feet of the human body, perform threshold segmentation on the obtained projection image, and obtain its maximum circumscribed rectangle, such as Figure 4 As shown, the bone region image is obtained by cutting out;

[0077] In the present invention, the OTSU threshold segmentation algorithm is used to perform threshold segmentation on the region of interest obtained in S22.

[0078] The bone region image in the image acquired by the present invention can be used as the basis for the subsequent graph cut algorithm.

[0079] More specifically, after obtaining the bone region image, the present invention can also expand it by 10 pixels on the basis of the image, but it cannot exceed the range of the image obtained by S1. The area outside this area is determined as a non-valid bone area, providing prior knowledge for subsequent graph cutting.

[0080] S3, performing threshold segmentation on the bone region image obtained in S2 to obtain the mask corresponding to the bone, based on which, a graph cut algorithm is used to segment each bone therein, that is, to obtain a bone segmentation image;

[0081] In the present invention, the bone region image obtained by S2 is segmented using the Yen histogram threshold to obtain the mask corresponding to the bone, such as Figure 5 shown.

[0082] Specifically, the bones obtained by segmenting using the graph cut algorithm include:

[0083] S31, creating a priori graph model according to the mask corresponding to the obtained skeleton;

[0084] Fill the closed contour of the mask corresponding to the obtained bone, refer to Figure 6 , take it as the determined foreground area, and mark the voxel value of the voxel therein as 2; according to S2, the area outside the circumscribed rectangle of the mask corresponding to the skeleton obtained in S22 is taken as the determined background area, and the voxel value of the voxel therein is marked as 1; the other areas in the mask corresponding to the skeleton are taken as unknown areas, and the voxel value of the voxel therein is marked as 0;

[0085] This creates a priori graphical models, such as Figure 7 As shown, S represents the foreground, T represents the background, and undirected edges are established between the elements in the graph model, that is, between the voxels of the input image. The weight of the adjacent voxel p to q is Among them, I p ,I q Respectively represent the voxel values ​​of voxel p and voxel q; σ 2 represents the variance of the voxel values ​​between all voxels in the graph model and their adjacent voxels; in this embodiment, Among them, I k I represents the voxel value of the kth voxel in the graph model; c It represents the voxel value of the voxel adjacent to the k-th voxel in the graph model. In the present invention, the voxel adjacent to the k-th voxel is a voxel selected from each of the up-down, left-right, and front-back directions corresponding to the voxel, for a total of three voxels as the voxels of its neighborhood; edge represents the number of edges in the graph model.

[0086] A background histogram is established for the voxels with a marked voxel value of 1, and a foreground histogram is established for the voxels with a marked voxel value of 2; wherein, the voxels with a marked voxel value of 1 are determined as background voxels in the image, and their weight to S is set to infinity, and their weight to T is set to 0; the voxels with a marked voxel value of 2 are determined as foreground voxels in the image, and their weight to S is set to 0, and their weight to T is set to infinity; the voxels with a marked voxel value of 0 are determined as voxels in other unknown areas of the image, and their weight to S is set to -λ*Ln(I / Hist B ), the weight to T is -λ*Ln(I / Hist o ), where λ is the weight coefficient, which is set to 0.01 in this embodiment; I is the voxel value of the current voxel; Hist O The weight of the voxel with a marked voxel value of 0 to the foreground voxel, Hist B is the weight of the voxel with a marked voxel value of 0 to the background voxel, which is calculated according to the above weight formula.

[0087] Thus, the energy E(L) of the graph can be obtained:

[0088]

[0089] Among them, R p (l p ) represents the region term, which indicates the weight of voxel p to T or S, where P represents all voxels in the image;

[0090] S32, solving the graph model by using a minimum cut or maximum flow algorithm to minimize the energy E(L) of the graph model, thereby obtaining an accurate bone segmentation result, that is, segmenting to obtain a connected domain corresponding to each bone in the bone region image;

[0091] In the present invention, after the connected domains corresponding to the bones in the skeletal region image are obtained, they can also be screened. Specifically, the connected domains with the largest area among the connected domains corresponding to the bones are calculated and obtained, and the connected domains corresponding to the areas larger than 0.5 times of the largest area are retained, thereby eliminating the interference of the clothes around the bones in the image, and finally segmenting the bones in the skeletal region image. Figure 8 shown.

[0092] In the present invention, before using the graph cut algorithm to segment each bone, the window position and width of the mask corresponding to the bone can be adjusted first. In this embodiment, the bone window can be set to [-400, 850].

[0093] S4, calculate the distance between each point in the bone segmentation image obtained in S3 and its nearest boundary, and obtain a distance map, such as Fig. 9 As shown; according to the characteristics of the bone area in the joint, the seed area of ​​each bone is obtained, and the seed area label map is marked, and the bone segmentation image obtained by S3 is semantically segmented by the watershed algorithm;

[0094] In the present invention, points with negative distances are marked as internal points, and points with positive distances are marked as external points.

[0095] Among them, the seed area label map is obtained, which specifically includes:

[0096] (1) Find the joint gap position in the bone segmentation image obtained by S3;

[0097] The bone segmentation image obtained by S3 is projected with maximum density along the left and right directions of the human body to obtain the corresponding two-dimensional image, and the two-dimensional image is segmented by threshold value to obtain a binary image to distinguish the foreground area and the background area. For example, Fig.10 As shown in (a), the white bright part is the foreground area and the black part is the background area;

[0098] By accumulating the pixel values ​​of the pixels in the same column along the front-back direction of the human body in the aforementioned binary image, the corresponding pixel value accumulation value curve can be obtained, such as Fig.10 As shown in (b), the average value of the pixel values ​​of all pixels in the binary image is taken as the reference value, as shown in Fig.10 As shown by the red line in (b), two areas where the cumulative values ​​of two continuous and smooth pixel values ​​are lower than the reference value are the backbone areas of the bones on both sides of the joint of the human limb, and the area between the two areas is the joint area.

[0099] The pixel points in the column corresponding to the minimum pixel value cumulative value in the pixel value cumulative value curve in the joint area are obtained, and the pixel points in this column are the corresponding joint gap positions.

[0100] (2) according to the calculated distance map, multiplying the minimum value of the distance therein by a set coefficient to obtain a set distance threshold, segmenting the aforementioned bone segmentation image according to the set distance threshold, and obtaining a plurality of connected domains through connected domain analysis;

[0101] In the present invention, the minimum value of the distance is the distance from the innermost center point to the nearest boundary in the skeleton segmentation image.

[0102] In the present invention, the setting coefficient can be 0.3.

[0103] (3) Calculate the size of each connected domain in step (2), determine the corresponding bone region according to the joint gap position obtained in step (1), and mark it as the corresponding seed region, that is, obtain the seed region label map, such as Fig.11 shown.

[0104] In the present invention, the size of the connected domain is specifically the volume occupied by all voxels therein.

[0105] In this embodiment, the knee joint is taken as an example, and the skeleton segmentation image obtained by S3 is semantically segmented by using the watershed algorithm. Fig.12 As shown, when determining the corresponding bone area, the direction of the human head and feet is taken as the standard, the direction of the human head is upward, and the direction of the human feet is downward. Then, in each connected domain obtained in step (2):

[0106] The connected domain located above the joint gap and with the largest area in the cross section is the femoral region, which is marked as the femoral seed region;

[0107] In the remaining part, the connected domain with the largest area on the cross section and with more than 100 pixels located above the joint gap is regarded as the patella region and marked as the patella seed region;

[0108] In the remaining part, located below the joint gap, there are two circumscribed contours in the cross section. The bone area corresponding to the smaller contour is the fibula area, which is marked as the fibula seed area; the bone area corresponding to the larger contour is the tibia area, which is marked as the tibia seed area.

[0109] In the present invention, the segmentation result obtained is as follows Fig.13 shown.

[0110] The present invention uses threshold segmentation to provide prior foreground and background knowledge based on the graph cut algorithm, obtains the mask of the bone part through calculation, and uses the anatomical relationship to establish the femur, tibia, and fibula seed area, and uses the watershed algorithm for segmentation. If the threshold method or the graph cut method is directly used, as long as there is adhesion, it cannot be effectively separated. The present invention first segments the foreground bone area, and then the watershed based on the distance map can be segmented by growth, so that the adhesion area can be disconnected, and the segmentation of the joint image can be achieved, and there will be no defects such as holes in the segmented image.

[0111] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention (including the claims) is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0112] The embodiments of the present invention are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the present invention.

Claims

1. A joint image segmentation method, characterized in that: include: S1, obtaining an image containing a joint to be segmented, determining whether there is metal therein, and removing the metal area therein if there is metal; S2, obtaining a bone region image from the image obtained in S1; S3, performing threshold segmentation on the bone region image obtained in S2 to obtain the mask corresponding to the bone, based on which, a graph cut algorithm is used to segment each bone therein, thereby obtaining a bone segmentation image; S4, calculating the distance between each point in the bone segmentation image obtained in S3 and its nearest boundary to obtain a distance map; According to the characteristics of the bone region in the joint, the seed region of each bone is obtained, and the seed region label map is obtained by marking; The skeleton segmentation image obtained by S3 is semantically segmented using the watershed algorithm.

2. The joint image segmentation method according to claim 1, characterized in that: In S3, the step of segmenting the skeleton using a graph cut algorithm includes: S31, creating a priori graph model according to the mask corresponding to the obtained skeleton; S32, solving the graph model by using a minimum cut or maximum flow algorithm to obtain a bone segmentation result, that is, segmenting each bone in the bone region image.

3. The joint image segmentation method according to claim 2, characterized in that: In S31, the graph model is specifically as follows: Fill the closed contour of the mask corresponding to the bone on the corresponding view, use it as the foreground area, and mark the voxel value of the voxel therein as 2; According to S2, the area outside the bone area of ​​the mask corresponding to the bone on the corresponding view is used as the determined background area, and the voxel value of the voxel therein is marked as 1; The other areas in the mask corresponding to the bones on the corresponding view are regarded as unknown areas, and the voxel value of the voxels in the marker is 0; Thus, a priori graph model is created, where S represents the foreground, T represents the background, and undirected edges are established between the elements in the graph model; For voxels with a marked voxel value of 1, set their weight to S to infinity and their weight to T to 0; for voxels with a marked voxel value of 2, set their weight to S to 0 and their weight to T to infinity; for voxels with a marked voxel value of 0, set their weight to S to -λ*Ln(I / Hist B ), the weight to T is -λ*Ln(I / Hist O ), where λ is the set weight coefficient, I is the voxel value of the current voxel, Hist O The weight of the voxel with a marked voxel value of 0 to the foreground voxel, Hist B is the weight of the voxels with a marked voxel value of 0 to the background voxels.

4. The joint image segmentation method according to claim 1, characterized in that: In S3, after the graph cut algorithm is used to segment the bones, connected domains can be screened, as follows: the connected domain with the largest area among the connected domains corresponding to the bones is calculated and obtained, and the connected domains corresponding to the areas larger than 0.5 times the maximum area are retained as the bones in the bone region image obtained by the final segmentation.

5. The joint image segmentation method according to claim 1, characterized in that: In S4, the marking obtains a seed region label map, which is as follows: (1) Find the joint gap position in the bone segmentation image obtained by S3; (2) according to the distance map obtained by calculation, the minimum value of the distance therein is multiplied by a set coefficient to obtain a set distance threshold, the bone segmentation image is segmented according to the set distance threshold, and a plurality of connected domains are obtained through connected domain analysis; (3) Calculate the size of each connected domain in step (2), determine the corresponding bone region according to the joint gap position obtained in step (1), and mark it as the corresponding seed region, thereby obtaining a seed region label map.

6. The joint image segmentation method according to claim 5, characterized in that: In the step (1), the step of finding the joint gap position in the bone segmentation image is as follows: Performing maximum density projection on the bone segmentation image along the left-right direction of the human body to obtain a corresponding two-dimensional image, and performing threshold segmentation on the two-dimensional image to obtain a binary image; The pixel values ​​of the pixel points in the same column along the front-back direction of the human body in the binary image are accumulated to obtain a corresponding pixel value accumulation value curve, and the average value of the pixel values ​​of all the pixel points in the binary image is used as a reference value, and the area in which the accumulated values ​​of two continuous and smooth pixel values ​​are lower than the reference value is obtained, which is the joint area; The pixel points in the column corresponding to the minimum pixel value cumulative value in the pixel value cumulative value curve in the joint area are obtained, and the pixel points in this column are the corresponding joint gap positions.

7. The joint image segmentation method according to claim 1, characterized in that: The S2 includes: S21, perform maximum density projection on the image obtained in S1 along the direction of the head and feet of the human body, perform threshold segmentation on the obtained projection image, and then perform connected domain operation to obtain several connected domains, obtain the circumscribed rectangle of the connected domain with the largest area as a bounding box, and use the bounding box to select the image obtained in S1 to obtain the region of interest in the image; S22, performing maximum density projection on the region of interest obtained in S21 along the direction of the head and feet of the human body, performing threshold segmentation on the obtained projection image, obtaining its maximum circumscribed rectangle, and cropping the image of the bone region therein based on the maximum circumscribed rectangle.

8. The joint image segmentation method according to claim 7, characterized in that: In S21, the threshold segmentation of the projection image of the image is performed using a moments histogram threshold segmentation algorithm.

9. The joint image segmentation method according to claim 7, characterized in that: In S22, the projection image of the region of interest is threshold segmented using the OTSU threshold segmentation algorithm.

10. The joint image segmentation method according to claim 1, characterized in that: In S1, the determination of whether there is metal in the image of the joint to be segmented is as follows: After using threshold segmentation to segment the cross-sectional, sagittal and coronal views of the image respectively, the mean Hu value of the foreground area is statistically obtained. If the mean is greater than the set threshold, it is considered that there is metal in the image, otherwise there is no metal in the image.

11. The joint image segmentation method according to claim 1, characterized in that: In S1, the metal area in the image containing the joint to be segmented is removed, specifically: The Hu values ​​of the points in the metal area of ​​the image containing the joint to be segmented are replaced with the set value.

Images